Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75703.13066892742,
-      "required_pages_per_second_upper_estimate": 0.006561755498552089
+      "required_token_positions_per_second": 75618.69360870571,
+      "required_pages_per_second_upper_estimate": 0.006554436708175845
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9462.891333615928,
-      "required_pages_per_second_upper_estimate": 0.0037748888358129598
+      "required_token_positions_per_second": 9452.336701088214,
+      "required_pages_per_second_upper_estimate": 0.003770678435091836
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18925.782667231855,
-      "required_pages_per_second_upper_estimate": 0.004378146737665323
+      "required_token_positions_per_second": 18904.673402176428,
+      "required_pages_per_second_upper_estimate": 0.004373263480705126
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28388.67400084778,
-      "required_pages_per_second_upper_estimate": 0.0017287213027901355
+      "required_token_positions_per_second": 28357.010103264645,
+      "required_pages_per_second_upper_estimate": 0.001726793137554954
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28388.67400084778,
-      "required_pages_per_second_upper_estimate": 0.0021854387094645523
+      "required_token_positions_per_second": 28357.010103264645,
+      "required_pages_per_second_upper_estimate": 0.0021830011349773244
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28388.67400084778,
-      "required_pages_per_second_upper_estimate": 0.3524923203105129
+      "required_token_positions_per_second": 28357.010103264645,
+      "required_pages_per_second_upper_estimate": 0.3520991606747786
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.641841748124989,
-    "producer_p95_seconds": 0.3580939550010953,
-    "gpu_update_seconds": 0.9739095245933057,
-    "producer_batches_per_second": 1.5580164470779565,
-    "gpu_batches_per_second": 1.0267894242204783,
-    "producer_headroom": 1.517367057282244,
+    "producer_batch_seconds": 0.31450038247686507,
+    "producer_p95_seconds": 0.3640840939988266,
+    "gpu_update_seconds": 0.9749970077704695,
+    "producer_batches_per_second": 3.179646371570187,
+    "gpu_batches_per_second": 1.0256441732951622,
+    "producer_headroom": 3.1001456980491624,
     "sustainable": true,
-    "buffer_seconds": 60.38239052478495,
+    "buffer_seconds": 60.44981448176911,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 238775.77894578734,
-    "upload_bytes_per_second": 5063752.244111188,
+    "generation_bytes_per_second": 235221.14753793247,
+    "upload_bytes_per_second": 5063403.408782123,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 51.23572374611106,
+    "estimated_full_shards_per_hour": 50.47298259356957,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 2.274885417002224,
-      "bytes": 15797221,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 3.0481821479988866,
       "bytes": 2407311,
@@ -562,6 +556,12 @@
       "bytes": 13189069,
       "pages": 31,
       "shards": 4
+    },
+    {
+      "seconds": 1.216554687001917,
+      "bytes": 10409429,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -330,3 +330,4 @@
 {"step": 270100, "loss": 0.6445578038692474, "lr": 0.0003, "grad_norm": 0.2480123810749485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39534484321, "tokens_per_second": 128718.24655918863, "data_wait_s": 1.6833000699989498e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 270200, "loss": 0.5957466680556536, "lr": 0.0003, "grad_norm": 0.15771897321231884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39549168780, "tokens_per_second": 149595.2218584798, "data_wait_s": 1.2221003999002278e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 270300, "loss": 0.6106796804815531, "lr": 0.0003, "grad_norm": 0.14322447692362245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39563684915, "tokens_per_second": 120176.25725778555, "data_wait_s": 1.2378004612401128e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 270400, "loss": 0.5690526589751244, "lr": 0.0003, "grad_norm": 0.14620697842648134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39578450717, "tokens_per_second": 149824.5612563105, "data_wait_s": 1.4336997992359102e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d276e5d9ad6f9984f10f1781e64ce5534b2e617222c204666f85f242643bc32
-size 708822
+oid sha256:5b42721b77db76f1db8ed1cb4ee462805545701ae1931f593b44d6c66565eee9
+size 710940